In 2006, data centers in the United States (U.S.) accounted for about 1.5% (about $4.5 billion) of the total electricity consumed in the U.S. This data center electricity consumption is expected to double by 2011. More than one-third of data center electricity consumption is for cooling servers, which could equate to more than about 1% of all U.S. electricity consumed by 2011. Electricity, personnel, and construction costs continue to increase and server hardware costs are decreasing, making the overall cost of cooling a large and growing part of the total cost of operating a data center.
The term “data center” (also sometime referred to as a “server farm”) loosely refers to a physical location housing one or “servers.” In some instances, a data center can simply comprise an unobtrusive corner in a small office. In other instances, a data center can comprise several large, wharehouse-sized buildings enclosing tens of thousands of square feet and housing thousands of servers. The term “server” generally refers to a computing device connected to a computing network and running software configured to receive requests (e.g., a request to access or to store a file, a request to provide computing resources, a request to connect to another client) from client computing devices, includes PDAs and cellular phones, also connected to the computing network. Such servers may also include specialized computing devices called network routers, data acquisition equipment, movable disc drive arrays, and other devices commonly associated with data centers.
Typical commercially-available servers have been designed for air cooling. Such servers usually comprise one or more printed circuit boards having a plurality of electrically coupled devices mounted thereto. These printed circuit boards are commonly housed in an enclosure having vents that allow external air to flow into the enclosure, as well as out of the enclosure after being routed through the enclosure for cooling purposes. In many instances, one or more fans are located within the enclosure to facilitate this airflow.
“Racks” have been used to organize several servers. For example, several servers can be mounted within a rack, and the rack can be placed within a data center. Any of various computing devices, such as, for example, network routers, hard-drive arrays, data acquisition equipment and power supplies, are commonly mounted within a rack.
Data centers housing such servers and racks of servers typically distribute air among the servers using a centralized fan (or blower). As more fully described below, air within the data center usually passes through a heat exchanger for cooling the air (e.g., an evaporator of a vapor-compression cycle refrigeration cooling system (or “vapor-cycle” refrigeration), or a chilled water coil) before entering a server. In some data centers, the heat exchanger has been mounted to the rack to provide “rack-level” cooling of air before the air enters a server. In other data centers, the air is cooled before entering the data center.
In general, electronic components of higher performing servers dissipate correspondingly more power. However, power dissipation for each of the various hardware components (e.g., chips, hard drives, cards) within a server can be constrained by the power being dissipated by adjacent heating generating components, the airflow speed and airflow path through the server and the packaging of each respective component, as well as a maximum allowable operating temperature of the respective component and a temperature of the cooling air entering the server as from a data center housing the server. The temperature of an air stream entering the server from the data center, in turn, can be influenced by the power dissipation and proximity of adjacent servers, the airflow speed and the airflow path through a region surrounding the server, as well as the temperature of the air entering the data center (or, conversely, the rate at which heat is being extracted from the air within the data center).
In general, a lower air temperature in a data center allows each server component to dissipate a higher power, and thus allows each server to dissipate more power and operate at a level of hardware performance. Consequently, data centers have traditionally used sophisticated air conditioning systems (e.g., chillers, vapor-cycle refrigeration) to cool the air (e.g., to about 65° F.) within the data center for achieving a desired performance level. By some estimates, as much as one watt can be consumed to remove one watt of heat dissipated by an electronic component. Consequently, as energy costs and power dissipation continue to increase, the total cost of cooling a data center has also increased.
In general, spacing heat-dissipating components from each other (e.g., reducing heat density) makes cooling such components less difficult (and less costly when considering, for example, the cost of cooling an individual component in a given environment) than placing the same components placed in close relation to each other (e.g., increasing heat density). Consequently, data centers have also compensated for increased power dissipation (corresponding to increased server performance) by increasing the spacing between adjacent servers.
In addition, large-scale data centers have provided several cooling stages for cooling heat dissipating components. For example, a stream of coolant, e.g., water, can pass over an evaporator of a vapor-compression refrigeration cycle cooling system and be cooled to, for example, about 44° F. before being distributed through a data center for cooling air within the data center.
The power consumed by a chiller can be estimated using information from standards (e.g., ARI 550/590-98). For example, ARI550/590-98 specifies that a new centrifugal compressor, an efficient and common compressor used in high-capacity chillers, has a seasonal average Coefficient-of-Performance (“COP”) from 5.00 to 6.10, depending on the cooling capacity of the chiller. This COP does not include power consumed by an evaporative cooling tower, which can be used for cooling a condenser in the refrigeration cycle cooling system and generally has a COP of 70, or better. The combined COP for a typical system is estimated to be about 4.7.
According to some estimates, some state-of-the-art data centers are capable of cooling only about 150 Watts-per-square-foot, as opposed to cooling the more than about 1,200 Watts-per-square-foot that could result from arranging servers to more fully utilize available volume (e.g., closely spacing servers and racks to more fully utilize floor-to-ceiling height and floor space) within existing data centers. Such a low cooling capacity can significantly add to the cost of building a data center, since data centers can cast as much as about $250 per-square-foot to construct.
As the air-cooling example implies, commercially available methods of cooling have not kept pace with increasing server and data-center performance needs, or the corresponding growth in heat density. As a consequence, adding new servers to existing data centers has become difficult and complex given the effort expended to facilitate additional power dissipation, such as by increasing an existing data center's air conditioning capacity.
Various alternative approaches for cooling data centers and their servers, e.g., using liquid cooling systems, have met with limited success. For example, attempts to displace heat from a microprocessor (or other heat-generating semiconductor-fabricated electronic device component, collectively referred to herein as a “chip”) for remotely cooling the chip have been expensive and cumbersome. In these systems, a heat exchanger or other cooling device, has been placed in physical contact (or close physical relation using a thermal-interface material) with the package containing the chip. These liquid-cooled heat exchangers have typically defined internal flow channels for circulating a liquid internally of a heat exchanger body. However, component locations within servers can vary from server to server. Accordingly, these liquid-cooling systems have been designed for particular component layouts and have been unable to achieve large-enough economies of scale to become commercially viable.
Research indicates that with state-of-the-art cooling, PUEs (as defined on page 10 hereinafter) of 1.4 might be attainable by 2011. However the costs to capitalize such cooling were not mentioned, and indicators suggest that saving electricity requires expensive equipment.
Immersion cooling of electronic components has been attempted in high-performance (e.g., computer gaming) applications, but has not enjoyed widespread commercial success. Previous attempts at immersion cooling has submerged some, and in some instances all, components mounted to a printed circuit board in a dielectric fluid using a hermetically sealed enclosure to contain the fluid. Such systems have been expensive, and offered by a limited number of suppliers. Large scale data centers generally prefer to use “commoditized” servers and tend to not rely on technologies with a limited number of suppliers.
Control systems have been used to increase cooling rates for a plurality of computers in response to increased computational demand. Even so, such control systems have controlled cooling systems that dissipate heat into the data center building interior air (which in turns needs to be cooled by air conditioning), or directly use refrigeration as a primary mode of heat dissipation. Refrigeration as a primary mode of cooling, directly or indirectly, requires significant amounts of energy.
Two-phase cooling systems have been attempted, but due to technical complexity, they have not resulted in cost-effective products or sufficiently low operating costs to justify investing in two-phase-cooling capital. Still other single- and two-phase cooling systems bring the coolant medium to an exterior of the computer, but reject heat to a cooling medium (e.g., air) external to the computer and within the data center (e.g., within a server room). Accordingly, each method of server or computer cooling currently employed or previously attempted have been prohibitively expensive and/or insufficient to meet increasing cooling demands of computing devices.
Indirectly, many researchers have tried to reduce the power of individual components such as the power supply and CPU. Although chips capable of delivering desirable performance levels while operating at a lower relative power have been offered by chip manufacturers, such chips have, to date, been expensive. Consequently, cooling approaches to date have resulted in one or more of a high level of electricity consumption, a large capital investment and an increase in hardware expense.
Therefore, there exists the need for an effective, efficient and low-cost cooling alternative for cooling electronic components, such as, for example, rack-mounted servers.